Generalizing from the results of randomized studies of treatment: Can non-randomized studies be of help?

Excerpt

Different patients who have the same illness may receive different forms of treatment. By documenting their treatment and the progression and complications of their illness, an attempt can be made to draw inferences regarding the relative impact of these different treatments. However, the interpretation of such comparisons can be compromised by underlying differences in the likelihood of the various outcome events among the patient treatment groups. Differences of opinion exist with respect to the frequency and magnitude of the confounding that may arise from these underlying differences. Benson and Hartz [1] concluded that “misuse of observational studies [of the efficacy of therapy] does not often occur in the recent literature”, and that nonrandomized studies of therapy “usually do provide valid information.” On the other hand, Pocock and Elbourne [2] state that greater reliance on the results of nonrandomized studies of therapeutic efficacy could lead to “considerable dangers to clinical research and even to the well-being of patients.” Particularly strident on this point was an editorial [3] which opined that “In an era of evidence-based medicine, controlled clinical trials are required to introduce new therapies into the clinic.” …